基于深度学习的粒子场数字全息成像研究进展

粒子场的数字全息成像中，由一幅粒子场全息图重建出高精度的三维粒子场分布，是数字全息技术领域的经典问题之一。相比于传统反向重建算法，深度学习算法可以从单个全息图直接重建出三维粒子场来简化算法复杂度，提高计算效率和准确率。介绍国内外研究团队将深度学习算法结合数字全息技术实现粒子场数字全息成像的研究进展，从不同粒子表征方法入手，叙述了支持向量机、全连接神经网络、全卷积网络、U-Net网络、深度神经网络在粒子场数字全息成像中粒子表征及粒子场反向重建过程中的应用原理、实现途径和准确率。最后指出了深度学习算法在这一研究领域的优势及目前基于深度学习算法的不足，并对如何进一步提高该方法的准确率进行了展望。     

雾化场光学全息与数字全息联合测量

针对雾化场光学全息测量存在干板湿化学处理繁琐、再现像采集耗时的问题,提出光学全息与数字全息联合测量的方法,建立由同轴光学全息、同轴数字全息以及数字延迟信号发生器组成的测量系统,并以双孔直射式喷嘴产生的雾化场为测量对象,利用该测量系统在一次测量中同时获得雾化场的光学全息和数字全息的再现图像,两者具有很好的一致性。光学和数字再现图像相对应的视场范围分别为27.87 mm4.77 mm和27.59 mm6.67 mm,数字方式获得视场范围内单一层面再现像的时间仅为8 s,而光学方式将近1 h。结果表明,光学全息与数字全息联合测量时,通过数字全息的雾化再现图像能够对实验总体效果进行实时评估,提高了雾化场全息测量的实验效率。     

同轴全息术用于粒子场测量的数值模拟

 采用数值方法模拟了同轴全息术测量粒子场的过程，对两种不同的数值算法＿直接傅里叶变换算法和卷积算法，进行了分析和比较，结果表明卷积算法符合实际要求。分析了记录图像的空间频谱及其对图像采样频率的要求，得出了在记录波长、采样间隔等条件一定的情况下的最小记录距离。对于一幅512×512像素的数字图像，若像元尺寸为6.7 μm，所用光波长为532 nm，则最小记录距离为43.2 mm。在此基础上对实验记录的振幅和相位型静态粒子的数字全息图，均得到了满意的数值再现像。     

Digital Holography for Instantaneous Spray Diagnostics on a Plane

We present a measurement technique that is capable of simultaneously determining sizes and positions of multiple transparent droplets in a plane from scattered light features. The technique is largely independent of particle intensity and mutual obscuration. Reflected and refracted light from the droplets in a pulsed laser sheet is recorded holographically to yield the smallest possible probe volume and the largest possible number density. Larger droplets are best analyzed at the image plane; in this case, the droplets appear as two spots (glare points), whose separation is proportional to the droplet diameter. Smaller droplets are easier to analyze at an out‐of‐focus plane, where their images appear as fringe patterns whose spatial frequency is related to droplet size. Photographic techniques allow only one of these planes to be chosen and are therefore not suitable for multidisperse sprays. Optical holography allows to analyze arbitrary depths, but often suffers from low sensitivity and long analysis times. With digital holography, the spray images are captured digitally by a CCD camera and reconstructed numerically; as in optical holography, the particle reconstruction plane can be freely chosen a posteriori to optimize the measurement. We discuss the issues raised by the transition from holographic film to a CCD sensor as the recording medium, and demonstrate the capabilities of the digital technique.     

数字同轴全息颗粒场检测中的颗粒分割识别与独立聚焦

针对数字同轴全息颗粒场检测中不同层面颗粒的自动聚焦问题,通过分析颗粒场数字同轴全息记录与再现的特点,提出一种颗粒分割识别与独立聚焦方法.给出了对单个颗粒进行聚焦判断的独立聚焦判据函数、以及实现颗粒分割识别和独立聚焦的方法.通过单层单个、两层两个及多层多个颗粒场的实验,验证了所提出方法的正确性.     

压缩感知重建数字同轴全息

数字重建是数字全息技术的关键步骤.传统的重建算法存在共轭像、聚焦物体与背景离焦物体相互干扰等问题.应用新兴的压缩感知技术,研究了全息图像的稀疏重建.基于衍射的线性运算,导出了利用压缩感知重建数字同轴全息三维空间的算法.利用该算法对颗粒的模拟全息图和数字显微全息实验全息图进行了重建,并将重建结果与传统的卷积重建结果进行了...     

All wavelet analysis of in-line particle holograms

In-line particle holography is a well-known method used for visualization and measurement of small objects such as sizes, 3-D positions and velocities. However, this method suffers from a major drawback that is conventional analysis of its reconstructed images is very tedious and time consuming. In this paper, we review recent progress in holographic particle tracking and sizing by using wavelet transform. Recent experimental results of spherical object measurement are noted.     

像面数字全息的重建相位误差分析和改善

像面数字全息是数字全息技术中常用的测量和成像方式,它通常采用离散傅里叶变换和频率滤波的方法进行物光波的重建.本文讨论了这些算法对重建相位的影响.首先分析了频谱泄露对于相位误差的影响,结果表明当采样周期为整数时,重建相位误差很小,因此具有极高的相位重建精度;而当不满足整周期采样时,相位重建误差有了明显的增加.为了改善频谱泄露所引起的相位误差,采用Hanning函数对数字全息图进行了预处理,结果表明Hanning窗的加入能够有效地提高重建相位的准确程度.     

Determination of optical field generated by a microlens using digital holographic method

In the paper, application of the digital holographic method for full field characterization of the beam generated by microlenses is considered. For this goal, the laboratory setup was designed based on Mach-Zehnder interferometry with the additional reference channel. The beam generated by a microlens was imaged by an afocal system and intensity distributions or interferograms (holograms) were registered by CCD camera. The digital holography using one image allows us to determine microlens parameters, i.e., focal length, aberrations, and shape. The optimum conditions to determine the surface shape of a microlens using holographic method have been found. We compare obtained results with geometrical and interferometric measurements. We show the advantage of digital holography for a shape microlens determination (improved accuracy), aberrations, and focal length (characterization facility). Through optimum refocusing, the digital holography gives more precise shape. The paper is accompanied with computer simulations and the experimental measurement data for geometrical, interferometric, and holographic methods.     

Quantitative detection and compensation of phase-shifting error in two-step phase-shifting digital holography

Phase-shifting digital holographic technique is a powerful tool for the measurement of various physical parameters, such as object deformation and liquid or cell’s refractive index change. However, for an accurate measurement, phase-shifting error in the reference wave path is still a major issue. In this paper, three novel and simple algorithms are proposed to quantitatively detect and correct phase-shifting error for a pure phase object in two-step phase-shifting digital holography. Influence of phase-shifting error is illustrated, and the effectiveness of the proposed algorithms is demonstrated by numerical simulation results.     

彩色数字全息的非插值波面重建算法研究

彩色数字全息及多波长照明的数字全息检测研究中,避免插值误差的可变放大率波面重建是一个重要的研究内容.由于Fresnel衍射积分可以表示成Fourier变换及卷积两种形式,对应地存在两种波面重建算法：其一,将重建距离分为两段的衍射“接力”算法;其二,用球面波为重建波的卷积算法.文中对这两种算法进行理论分析及实验研究,讨论让重建计算满足取样定理的条件.结果表明,卷积算法较容易满足取样定理,能够获得较好的重建物光场.     

彩色数字全息的非插值波面重建算法研究

彩色数字全息及多波长照明的数字全息检测研究中,避免插值误差的可变放大率波面重建是一个重要的研究内容.由于Fresnel衍射积分可以表示成Fourier变换及卷积两种形式,对应地存在两种波面重建算法：其一,将重建距离分为两段的衍射“接力”算法;其二,用球面波为重建波的卷积算法.文中对这两种算法进行理论分析及实验研究,讨论让重建计算满足取样定理的条件.结果表明,卷积算法较容易满足取样定理,能够获得较好的重建物光场. 关键词： 彩色数字全息 波面重建 衍射计算     

基于希尔伯特-黄变换方法的同轴粒子全息图中粒子轴向位置提取方法

在颗粒的数字全息中,传统方法应用数值再现反演计算得到颗粒的相关信息,其中再现的判焦过程中存在繁琐费时的缺点.本文提出了一种基于希尔伯特-黄变换方法的同轴粒子全息图分析方法.将粒子同轴全息图沿中心往外得到的单个粒子径向强度分布作为初始信号,根据希尔伯特-黄变换方法中的经验模态分解首先将信号分解成几个本征模态函数,通过分析第一本征模态函数的希尔伯特谱,由拉依达法则剔除部分奇异点后做最小二乘线性回归分析,得到粒子的空间位置.该方法不需要对全息图进行重建,由信号自身构建基函数,有很强的适应性.理论模拟和实验证明,该方法计算速度快,准确度高,有望应用于全息图的实时在线分析.     

预成像数字全息测量大物体三维形貌

为了扩大数字全息三维形貌测量的视场,提出了一种预成像的方法.物体先被成像镜头成一缩小的实像,再使用离轴菲涅耳数字全息对此实像进行全息记录.采用倾斜照明光的方法使两次数字全息的再现像之间产生相位差,通过相位去噪和去包裹处理从相位差中提取物体的三维形貌数据.实验证明,预成像方法使测量视场的大小得到了至少一个数量级的提高,采用离轴菲涅耳数字全息降低了测量系统对环境稳定性的要求.     

Curvature measurement of optical surface using digital holography

In this paper, an optical surface curvature measurement technique based on digital holography is proposed. Unlike the previous digital holographic methods, which only involve a small part of the measured data for parameter estimation, the proposed method adopts more of them using least square method. Hence, the analyzed result is not affected by data extraction direction and less sensitive to imperfect phase compensation. The proposed method is demonstrated with a sample of reflection mirror in laser resonator, and verified by comparing the measurement result with that of the white light interferometer.     

High resolution digital holography

In digital Fresnel holography, present specifications of charge-coupled device cameras require the incident beams to be quasi-parallel. That implies large speckle grain size and low lateral resolution in reconstructed images. Better lateral resolutions are demonstrated in our work, down to 8 μm, allowing the observation of sub-millimetre objects by digital holography. The experimental set-up built samples the incident light distribution with a definition of 500 pixels mm⁻¹. The maximum acceptable angle is then widened and the hologram recording and reconstruction distances are drastically reduced, as well as the speckle size. Interferometric holography was implemented for the measurement of the deformation of a sub-millimetre silicon cantilever and results in phase-contrasts imaging are also reported. The design of dedicated complementary metal-oxide-semiconductor active pixel sensor cameras is also discussed.     

菲涅尔数字全息的脉冲响应

数字全息是用CCD记录全息图并用计算机数值重建全息像的一种全息新方法.在数字全息中,通过对不同记录参数下记录的全息图的数值处理,可以消除零级光和共轭光,从而将数字全息系统看作是一个线性系统.本文依据全息理论和付里叶频谱分析,对菲涅尔数字全息系统的脉冲响应和分辨本领进行了理论分析.结果表明,在矩形等间隔抽样的情形下,菲涅尔数字全息的脉冲响应是由CCD有限大小的孔径衍射斑调制的矩形函数;菲涅尔数字全息的分辨率由CCD的孔径尺寸决定;由于CCD像素具有一定的大小,使得点光源的像发生弥散.     

Particle digital in-line holography with spherical wave recording

In this paper, we propose a method of digital in-line holography of particle. A diverging spherical beam is used for illumination in recording hologram, the complex amplitude distribution generated by particle field at a single plane located in the Fresnel diffraction region is recorded by CCD, and a plane beam for reconstructing hologram, then, the magnified image can be obtained by numerical reconstruction in computer. This procedure can be interpreted by Fourier optical theory and the theoretical analysis have been done in detail, the experimental results, the air freshener being subject, are also given.     

Improvement of signal-to-noise ratio in digital holography using wavelet transform

The basic problem in optical and digital holography is the presence of speckle noise in the reconstruction process, which reduces the signal-to-noise ratio (SNR). The presence of speckle noise is serious drawback in optical and digital holography since it substantially reduces the SNR in the reconstructed image. In this paper, we present wavelet filtering to improve SNR in the reconstructed images from digital holograms. Experimental results are presented.     

基于滤波成像的大视角数字全息技术

为提高数字全息再现像视角,提出一种基于滤波成像的数字全息技术来实现大视角的三维物体面型测量.利用离轴像面数字全息技术,通过在4F相干图像处理系统的空间频谱面处放置可移动的低通滤波器,使满足CCD分辨率的物光波与参考光波干涉形成全息图,并控制低通滤波及成像区域分别记录不同谱段的子全息图.再现时,首先对子全息图进行数字傅里叶变换,重构对应频谱段,并对频谱段进行拼接形成完整的物光频谱|而后通过数字再现获得大视角的数字全息再现像.利用该方法测量了圆柱形表面（光滑的缝纫针）的三维形貌,并取得了较好的实验结果.